@aldus-palace/mcp@0.7.7

Patch Changes

• Updated dependencies [41ec277]
  • @aldus-palace/core@0.13.0

@aldus-palace/mcp@0.7.6

Patch Changes

• Updated dependencies [62a5eac]
  • @aldus-palace/core@0.12.0

@aldus-palace/mcp@0.7.5

Patch Changes

• Updated dependencies [b74a164]
• Updated dependencies [9b144f2]
• Updated dependencies [e18d505]
• Updated dependencies [0234daa]
  • @aldus-palace/core@0.11.0

@aldus-palace/mcp@0.7.4

Patch Changes

• Updated dependencies [d3c2d19]
• Updated dependencies [eedbd5a]
• Updated dependencies [e5254cd]
• Updated dependencies [3d87708]
• Updated dependencies [f9332f7]
  • @aldus-palace/core@0.10.0

@aldus-palace/mcp@0.7.3

Patch Changes

• Updated dependencies [8e1c285]
• Updated dependencies [118f319]
  • @aldus-palace/core@0.9.0

@aldus-palace/mcp@0.7.2

Patch Changes

• Updated dependencies [d0d813e]
• Updated dependencies [bde8de8]
  • @aldus-palace/core@0.8.0

@aldus-palace/mcp@0.7.1

Patch Changes

• Updated dependencies [28a9586]
  • @aldus-palace/core@0.7.0

@aldus-palace/mcp@0.7.0

Minor Changes

• ccff588: Expose the Action Gate: list_actions, decide_action and revoke_action,
  with an actions profile and the aldus-palace-mcp-actions binary. The tools
  return a human card with the full payload in structuredContent.
• 04d25f0: list_actions returns the trust score and autonomy level alongside the queue,
  in the card footer and in structuredContent.
• 7e481cb: capture and list_today return a short human card as text content, with the
  full payload in structuredContent and an output schema. The model quotes a
  receipt instead of reading a JSON wall.

Patch Changes

• Updated dependencies [747243d]
• Updated dependencies [4ea1c06]
• Updated dependencies [60768cf]
• Updated dependencies [1c253c9]
• Updated dependencies [ccff588]
• Updated dependencies [93201b4]
• Updated dependencies [dbe23fc]
• Updated dependencies [120bd72]
• Updated dependencies [8bdc95c]
• Updated dependencies [9a4d5e0]
• Updated dependencies [5bcbd6b]
• Updated dependencies [04d25f0]
  • @aldus-palace/core@0.6.0

@aldus-palace/core@0.9.2

Patch Changes

• 38175b8: Harden the FTS5 retriever and true deletion.

  • A query that contains an emoji or a punctuation-only word ("? hello",
    "😀 hello") no longer produces a dangling OR; the expression stays valid,
    so a capture never fails on such input.
  • Context retrieval falls back to the keyword pass when the FTS query errors,
    instead of failing the whole capture.
  • purgeUserData now also deletes the memory_search rows, so a purged memory
    leaves nothing behind in the full-text index.

@aldus-palace/core@0.9.1

Patch Changes

• 106c281: toMatchQuery OR-joins the words of a query (each Latin word a prefix term,
  each CJK word a character phrase) so a memory that matches any word is a
  candidate and bm25 ranks the rest. The retrieval benchmark now measures
  Recall@K and MRR (pnpm bench:retrieval).